Traffic Anomaly Detection Method Integrating Spatio-temporal Features Of Road Network

The transportation system is one of the most essential components to support a city.With the rise of data collection and Big Data technology,data plays a pivotal role in urban governance.At present,more and more cities are beginning to build intelligent transportation systems(ITS)to manage traffic with the help of data.Traffic anomaly refers to the abnormal event that occurs in the urban transportation system,which could have a potential negative impact on the normal operation of the urban transportation system.Therefore,if the traffic anomaly can be detected in the early stage,the impact could be avoided,which has positive significance in practical applications.This thesis focuses on the temporal and spatial anomalies in the road network system based on real-time traffic data.The temporal and spatial characteristics of the traffic in the inner ring area of Nanjing are explored.We combine existing methods to propose a traffic anomaly detection method that integrates the temporal and spatial features of the road network,which provides some guidance and reference to the city manager.The main research content and conclusions of this paper are as follows.First,we build a road traffic spatiotemporal database based on long-term collection of real-time traffic data in the inner ring area of Nanjing,and make a detailed analysis of the traffic spatiotemporal distribution.Using time series analysis to get the characteristics of traffic in time,it is found that in general,the road speed in the study area fluctuates periodically.The short daily period includes an obvious morning peak start from 7:00 to9:00 and an evening peak from 17:00 to 20:00.The weekly long period is made up with weekdays and weekends,which shows a different pattern in the morning peak.Microscopically,different types of roads have different temporal feature.Among them,the fluctuation of the urban expressway is more obvious than the urban branch road.In addition,we analyze the spatial autocorrelation of road speeds using the Moran’s I based on the improvement of the road network neighborhood.The result shows that the spatial aggregation of road speeds is related to the average road speed with a time periodicity.Second,with referring the existing methods,this paper proposes a basic definition of the traffic anomaly based on the characteristics of the data used in this article.The road network is abstracted as a directed graph,and the spatial feature of each road segment in the road network is extracted using the graph-embedding method.A spatiotemporal fusion neural network is designed to fit the general trend of the road speed,and local outlier factor(LOF)is used to detect from the residuals to locate the anomalies.Finally,a constructed anomaly data set is used to adjust the parameters of the model.The effectiveness of the method in the extraction of temporal and spatial anomalies in road networks can be judged from the accuracy indicators.The result shows our method performs the best on both constructed and real data sets compared with the existing methods.Meanwhile,it is verified that the use of graph embedding method can better extract the spatial information of the road.Third,there are many manifestations of anomalies in spatiotemporal data.Anomalies in some scenarios can often reflect some specific patterns.According to different scenarios,we propose three anomaly detection modes: temporal anomaly detection mode and spatial anomaly detection mode and spatiotemporal anomaly detection mode.The time series detection mode detects anomalies in the historical data of a certain road from the time dimension.The spatial anomaly detection mode can detect anomalies in some roads at a current time node from the space dimension.And the spatiotemporal anomaly detection mode can detect the wide-range anomalies,which is a shortage of the spatial anomaly detection mode.In actual cases,the model in this paper can better identify the anomalies caused by traffic accidents and freezing weather.